Comparing Alternative Psychiatric Diagnostic Criteria Sets Using Item Response Theory with Total Information Curves

• Activity Number: 190
• Type: Contributed
• Date/Time: Monday, August 1, 2011 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Melanie M. Wall*+ and Deborah S. Hasin
• Companies: Columbia University and Columbia University
• Keywords: item response theory ; latent variables ; measurement ; substance abuse

Abstract

IRT models are useful for assessing the ability of criteria sets to measure underlying psychiatric disorders, e.g. substance use disorders. IRT assumes the underlying psychiatric disorder is inherently continuous ranging from low to high severity and estimates how well each observable criterion measures different points along that continuum. The total information curve summarizes the combined criteria and provides a visual inspection of where along the disorder continuum the criteria measure the disorder well (i.e. with more precision). In the current work, we develop a way to statistically test whether the total information provided by one criteria set is larger than another. Standard errors for the total information are developed based on asymptotic theory as well as using the bootstrap procedure. We illustrate the method with an application to nicotine use disorder. With the changes proposed for the other DSM-5 substance use disorders to combine dependence and abuse criteria along with cravings into a single disorder, we examined whether combining nicotine dependence, abuse and cravings criteria is warranted.